基于DDC算法的稳健主成分法及其实证  

DDC Algorithm-based Robust PCA and Its Empirical Study

作　　者：阮皓麟 王斌会[1] Ruan Haolin;Wang Binhui(School of Management,Jinan University,Guangzhou 510632,China)

机构地区：[1]暨南大学管理学院,广州510632

出　　处：《统计与决策》2020年第16期36-40,共5页Statistics & Decision

基　　金：国家社会科学基金资助项目(16BTJ035)。

摘　　要：经典主成分分析对离群值非常敏感,常因离群值的存在导致分析结果偏离实际。针对含有离群值的数据,传统的稳健主成分分析一般通过识别并删除离群样本后建模达到稳健效果。然而,有些情况下离群样本中只有少数几个变量发生离群,剩余的大部分变量并无离群。删除离群样本的做法在这种情况下显然是不妥的。文章基于DDC算法提出一种稳健主成分分析法DDCPCA。模拟和实证结果表明,面对含该类离群样本的数据,该方法与传统稳健主成分分析法相比具有较大的优势。Classical principal component analysis(PCA) is so sensitive to outliers that the existence of outliers often leads to the deviation of analysis results from the reality. Traditional robust PCA methods achieve robust results by modelling after detecting and deleting outlying samples. However, sometimes there are just a few outlying variables in an outlying sample, while the rest are all normal variables, in which case deleting outlying samples is obviously not appropriate. This paper puts forward a robust PCA based on DDC(Detecting Deviating Cell) algorithm named DDCPCA(Detecting Deviating Cell Principal Component Analysis). Simulations and empirical study prove that the proposed algorithm is superior to the traditional robust PCA for data containing such outliers.

关 键 词：离群数据格 DDCPCA 稳健主成分分析 

分 类 号：C81[社会学—统计学]